Positive association and shared genetic structure between cardiovascular diseases and osteoarthritis: Insights from cross-sectional study and genome-wide pleiotropic analysis

Abstract

Background

Osteoarthritis (OA) and cardiovascular diseases (CVDs) are leading global causes of disability. While evidence indicates a significant association in their prevalence, the underlying biological mechanisms driving this comorbidity remain unclear.

Objectives

This study seeks to elucidate the epidemiological association between OA and common CVDs, and investigate whether shared genetic architecture underlies their comorbidity.

Design

cross-sectional study and genome-wide pleiotropic analysis.

Methods

Among 21,019 participants, multivariable logistic regression was used to examine the associations between OA and five CVDs (angina pectoris, coronary artery disease, myocardial infarction, heart failure, and stroke). Genetic correlations were then estimated using large-scale GWAS summary data of European ancestry, with the OA GWAS specifically focusing on knee and hip. Subsequent pleiotropy and multi-trait colocalization analyses were performed to identify shared loci, genes, and biological mechanisms.

Results

A positive association was observed between the prevalence of CVDs and OA, potentially explained by shared genetic factors. The genetic correlations (rg) and P values were: AP-OA (rg = 0.2229, p < 0.001), CAD-OA (rg = 0.1530, p < 0.001), MI-OA (rg = 0.1685, p < 0.001), HF-OA (rg = 0.3094, p < 0.001), and stroke-OA (rg = 0.1605, p < 0.001), indicating a robust and consistent shared genetic architecture between OA and these cardiovascular conditions. A total of 80 pleiotropic loci were identified, with 20 showing significant colocalization. Gene-level analysis revealed 84 distinct pleiotropic genes, including APOE, SMG6, UQCC1, and FES. Tissue enrichment analysis highlighted the role of pleiotropic mechanisms in central and peripheral blood vessels. Pathway assessment and multi-trait colocalization suggested that abnormal lipid metabolism may be a common cause of CVDs and OA.

Conclusion

Our study revealed a positive association and a shared genetic connection between CVDs and OA, and illuminated the potential mechanisms that could be involved.

Plain language summary

Both osteoarthritis and cardiovascular diseases are leading causes of disability worldwide. While previous epidemiological studies show a positive association in the prevalence between osteoarthritis and several cardiovascular diseases, the underlying reasons for this link have remained unclear. In this study, we confirmed a consistent positive relationship between osteoarthritis and five major cardiovascular conditions involving 21,019 participants, including angina pectoris, coronary artery disease, myocardial infarction, heart failure, and stroke. To understand why these two seemingly different types of diseases often occur together, we moved beyond lifestyle factors to analyze their large-scale genetic data, discovering a significant shared genetic architecture between them. Our analysis identified 80 specific genetic locations and 84 distinct genes (such as APOE and FES) that play dual roles in both joint and heart health. These shared genetic mechanisms appear to be particularly active in our blood vessels and suggest that abnormal lipid metabolism may be a common driving force behind both conditions. By uncovering these shared biological pathways, our research partially explains the genetic foundation of this comorbidity and points toward new therapeutic targets that could potentially treat both osteoarthritis and cardiovascular diseases.

Introduction

Osteoarthritis (OA) and cardiovascular diseases (CVDs) are two of the most prevalent health challenges facing the modern world. OA affects over 527.8 million individuals globally, standing as a leading cause of physical disability.^1,2 Particularly in weight-bearing joints like the knee and hip, OA is highly prevalent (global prevalence: 22.9% [95% CI: 19.8–26.1] for knee OA aged ≥40; 8.55% [95% CI: 4.85–13.18] for hip OA with K-L grade ≥2) and serves as the primary indication for total joint replacement, profoundly impacting quality of life and creating a heavy socioeconomic burden.^3–5 Meanwhile, CVDs, encompassing a spectrum of conditions such as coronary artery disease (CAD) and stroke, are the top contributors to global morbidity and mortality. ⁶ From 1990 to 2019, CVD prevalence nearly doubled, accompanied by a significant escalation in the disability-adjusted life years (DALYs), with ischemic heart disease alone accounting for 182 million DALYs in 2019. ⁷ Although people are increasingly aware of their devastating effects and give high priority to OA and CVDs, their pathogenesis remains unclear, which limits the optimization of clinical treatment options and the development of more effective therapeutic strategies.^8,9 Thus, it is essential to further clarify the mechanisms behind OA and CVDs to pave the way for developing effective therapeutic interventions.

Despite distinct clinical presentations and lesion sites, epidemiological studies show a significant association in the prevalence between OA and several CVDs. For instance, patients with self-reported OA face a 22% heightened risk for myocardial infarction (MI). ¹⁰ Consistently, another study found OA patients have a significantly elevated risk of heart failure (hazard ratio, HR=1.45) and transient ischemic attack (HR=1.55). ¹¹ Researchers deem this association is not mere a coincidence and propose several plausible explanations, such as overlapping risk factors including age and overweight, as well as shared pathways like inflammatory responses and lipid metabolism dysregulation.^12–14 Several studies based on Mendelian randomization also discuss whether there is a genetic or potential causal relationship between OA and CVDs,^15–17 yet these results are mixed and inconclusive, underscoring the need for further research to determine the mechanisms connecting these two conditions.

Until now, no studies investigate whether there is a shared genetic architecture promoting the co-occurrence of CVDs and OA, letting alone identify common susceptibility loci and genes. Recent advent of linkage disequilibrium score regression (LDSC) analysis has provided a valuable tool for assessing genetic correlations between different diseases. ¹⁸ However, the extent to which these correlations are driven by specific loci or the entire genome remains unclear. To compensate for this deficiency, we utilized Pleiotropic Analysis under Composite null hypothesis (PLACO), a novel cross-trait analysis method. This approach identifies pleiotropy at the SNP level based on a level-α intersection–union test (IUT) and has proven effective in detecting common loci across different diseases. ¹⁹ Consequently, it enables an in-depth exploration of the shared genetic mechanisms underlying OA and CVDs.

This study aimed to clarify the epidemiological associations between OA and common CVDs, and investigate their potentially shared genetic architecture. By leveraging large-scale GWAS summary statistics for knee and hip OA and five common CVDs (angina pectoris (AP), CAD, MI, heart failure (HF), and stroke) derived from predominantly European-ancestry populations, we sought to evaluate the genetic overlap between these conditions. Specifically, the ‘PLACO’ method was employed to identify potential pleiotropic risk loci and candidate genes, followed by functional annotation and tissue-specific analyses. This integrated framework holds the potential to uncover shared genetic mechanisms and identify novel therapeutic targets for the management of OA and CVDs.

Methods

2.1 Study design

This study employs a cross-sectional design integrated with a multi-stage genome-wide pleiotropic analysis to investigate the epidemiological associations and shared genetic architecture between OA and five CVDs. Clinical correlations were evaluated through multivariable logistic regression using cohort data of the English Longitudinal Study of Ageing (ELSA) and the National Health and Nutrition Examination Survey (NHANES), while genome-wide genetic correlations were estimated via linkage disequilibrium score regression (LDSC) and high-definition likelihood (HDL) analyses. Pleiotropic loci were identified using PLACO, and candidate genes were prioritized through multi-marker analysis of genomic annotation (MAGMA) and summary-based Mendelian randomization (SMR). The underlying biological mechanisms were further explored using tissue-specific and pathway enrichment analyses. Finally, Hypothesis prioritization for multi-trait colocalization (HyPrColoc) analyses were performed to characterize shared causal variants and investigate the potential mediating role of lipid metabolism in the comorbidity of CVDs and OA. The framework of the study design is presented in Figure 1.

Figure 1.

Study workflow.

2.2 Study population

The Ethics Committee of Xijing Hospital granted approval for the study (ethical approval number: KY20253289-1). And the reporting of this study conforms to the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement. ²⁰ Data for the cross-sectional component of this study were derived from the ELSA and the NHANES. ELSA is a comprehensive longitudinal study examining the health, well-being, and social circumstances of older adults in England which provides valuable insights into ageing and informs policies to enhance the quality of life for seniors. ²¹ NHANES database is a large, nationally representative, multi-stage, stratified study conducted on the US population that evaluates nutrition and health of the US population. ²² Both ELSA and NHANES obtained ethical approval from their respective institutional review boards, and written informed consent was obtained from all participants at the time of original data collection. As this study involved secondary analysis of de-identified, publicly available data, no additional informed consent was required per the data use agreements.

Participants were enrolled from Wave 8 (2016–2017) of the ELSA and seven cycles (2005–2006, 2007–2008, 2009–2010, 2011–2012, 2013–2014, 2015–2016, and 2017–2018) of the NHANES. The overall study period therefore spanned from 2005 to 2018. For the ELSA cohort, a total of 8,445 individuals were initially recorded. Participants were excluded based on the following criteria: (1) Based on the biological nature of OA as a degenerative disease that typically manifests after age 40, ²³ and consistent with the demographic composition of the databases (particularly ELSA with few data for groups <40 years old), participants aged <40 years were excluded to ensure sufficient statistical power; (2) missing arthritis data or diagnosed with non-osteoarthritis types; (3) missing data on cardiovascular outcomes (AP, MI, HF, and stroke); and (4) missing key covariates (including age, sex, white ethnicity, education, BMI, smoking, alcohol consumption, records of hypertension, and diabetes). After applying these criteria, 6,135 participants were included in the final analysis. Regarding the NHANES cohort, 70,190 participants were initially identified from the 2005–2018 cycles. We excluded those: (1) aged <40 years; (2) with missing arthritis data or non-osteoarthritis diagnoses; (3) with missing information on cardiovascular conditions (CAD, AP, MI, HF, and stroke); and (4) with missing essential covariates (including age, sex, race, education, PIR, BMI, smoking, alcohol consumption, records of hypertension, and diabetes) or sampling weights. Rather than predetermining the sample size via an a priori power analysis, we adopted a maximal-inclusion strategy. To maximize the statistical power and generalizability of our findings, all eligible participants who met the inclusion criteria from the ELSA and NHANES cohorts were enrolled. Following these rigorous exclusions, the final NHANES sample consisted of 14,884 participants. The process of participant selection based on inclusion and exclusion criteria is depicted in Figure 2.

Figure 2.

Flowchart of participant selection procedure of the ELSA cohort (A) and the NHANES cohort (B).

Abbreviations: OA, osteoarthritis; CVD, cardiovascular disease; AP, angina pectoris; CAD, coronary artery disease; MI, myocardial infarction; HF, heart failure.

2.3 Assessment of CVDs

The identification of CVD in both cohorts was based on self-reported physician diagnoses. In the ELSA study, participants were asked, “Has a doctor ever told you that you have heart disease, including angina, heart attack, congestive heart failure, or other heart problems?” and “Has a doctor ever told you that you have had a stroke?”; individuals reporting either condition were classified as having CVD. For the NHANES cohort, CVD was defined by positive responses to questions regarding AP, CAD, MI, HF, and stroke: “Has a doctor or other health professional ever told you that you have angina pectoris/coronary artery disease/heart attack (also called myocardial infarction)/heart failure/stroke?” Additionally, family history of early-onset heart disease was identified through the question, “Including living and deceased, were any of your close biological that is, blood relatives including father, mother, sisters or brothers, ever told by a health professional that they had a heart attack or angina before the age of 50?”

2.4 Assessment of OA

The identification of OA followed a similar two-step procedure in both cohorts. In the ELSA study, participants were first asked, “Have you ever had arthritis?” or “Has a doctor or other health professional ever told you that you have or have had arthritis?”; those who answered “yes” were subsequently asked the follow-up question, “Which type of arthritis do you have?” Similarly, in NHANES, OA status was determined through a two-step sequence: (1) “Has a doctor or other health professional ever told you that you had arthritis?” and (2) “What type of arthritis was it?”

Participants were assigned to the OA group if they answered ‘Yes’ to the first question and specifically identified OA as the only subtype in the follow-up question. Conversely, those who responded ‘No’ to the initial arthritis screening question were assigned to the Non-OA group, who served as the reference control population for comparative and regression analyses.

2.5 Covariates

Covariates for multivariate analysis were selected by assessing those used in previous studies with similar methodologies, including demographic information and self-reported medical history and health-associated behaviors both in the NHANES database and the ELSA database. The demographic variables included age, sex (male/female), race/ethnicity (non-Hispanic white, non-Hispanic black, other race in the NHANES study, white and non-white in the ELSA study), level of education, poverty income ratio (PIR) (<1.3, 1.3-3.5, ≥3.5 in the NHANES study). Weight (kg) divided by height squared (m²) was used to compute BMI in the NHANES study. Correspondingly, question “Given your age and height, would you say that you are about the right weight/too heavy/too light?” was asked to participants in the ELSA study to roughly categorize BMI. Health-associated behaviors encompassing alcohol intake (yes/no) and smoking. To investigate the history of NSAIDs use, the prescription medications questionnaire of NHANES was used. Self-reported medical history included a history of hypertension, diabetes and cardiovascular diseases.

2.6 Statistical analysis for ELSA and NHANES data

Continuous variables were presented as the mean ± standard deviation (SD) and compared using the T-test, while categorical variables were expressed as percentages (95% CI) and compared using the Rao-Scott chi-square test. For the ELSA data, multivariate logistic regression analysis was employed to evaluate the association between CVDs and OA, reporting odds ratios (ORs) with 95% CIs. Three adjustment models were utilized: Model 1 was unadjusted; Model 2 was adjusted for age, sex, and race; and Model 3 was further adjusted for education, BMI category, alcohol consumption, smoking behavior, hypertension, and diabetes. Regarding the NHANES data, weighted multivariate logistic regression analysis was conducted to assess the correlation between CVDs and OA. The adjustment models followed a similar structure: Model 1 was unadjusted; Model 2 was adjusted for age, sex, and race; and Model 3 was adjusted for Model 2 plus education, PIR, BMI, use of NSAIDS, alcohol consumption, smoking behavior, hypertension, and diabetes. In accordance with NHANES analytical guidelines, we applied the 2-year Mobile Examination Center (MEC) weights corresponding to the variables collected from the smallest respondent group to account for the complex survey design. Subgroup analyses stratified by age, sex, race, and BMI category were conducted, with interaction tests performed to assess the heterogeneity of correlations across subgroups.

2.7 GWAS data set collection

We obtained GWAS summary statistics from publicly accessible data sources of European ancestry (>99.0%), and selected GWAS studies with sample sizes exceeding 50000 to guarantee statistical power. Specifically, the knee and hip OA GWAS dataset (MSK-KP meta-GWAS) included 490,345 individuals (89,741 cases, 99.1% European); AP (pheweb meta-GWAS) included 470,931 European individuals (30,025 cases); CAD (GWAS catalog) included 1,165,690 European individuals (181,522 cases); MI (FinnGen) included 369,139 European individuals (26,060 cases); HF (GWAS catalog) included 977,350 European individuals (47,309 cases); and Stroke (GWAS catalog) included 1,308,460 European individuals (73,652 cases). The exclusion of SNPs located in the major histocompatibility complex (MHC) region was performed, namely the 25Mb to 35Mb interval on chromosome 6. SNPs with a minor allele frequency exceeding 0.01 were retained to concentrate on common genetic variants. Specific information was shown in Table S1.

2.8 Genetic overlap at the genome wide level

We employed both LDSC and HDL analyses to evaluate the genetic correlations across the genome for five distinct trait pairs of CVD and OA. For the LDSC analysis, we utilized the well-imputed HapMap3 variants. Additionally, we used the pre-computed LD scores specific to European ancestry derived from the 1000 Genomes Project Phase 3¹⁸. The HDL analysis was conducted using the R package HDL-v1.4.0, collecting 1,029,876 well-imputed HapMap3 SNPs as reference panel. ²⁴ For both LDSC and HDL analysis, we applied a Bonferroni-corrected significance threshold of P < 0.01 (0.05/5) to account for multiple testing across the five pairs of CVD-OA.

Furthermore, stratified-LDSC (S-LDSC) was applied to assess the statistical significance of the SNP heritability enrichment in 54 human tissue datasets from the Genotype-Tissue Expression (GTEx) project, which was conducted by examining the Z-scores of the regression coefficients and their corresponding P values. ²⁵

2.9 Determination and annotation of pleiotropic loci and genes

For the trait pairs showing significant genetic correlation, we utilized the recently introduced PLACO analysis to identify pleiotropic SNPs. ¹⁹ We considered SNPs with a PLACO value of less than 5 × 10^-8 as significant pleiotropic variants. And then we employed the Functional Mapping and Annotation (FUMA) of GWAS to ascertain the genomic regions harboring these risk variants, known as pleiotropic loci. ²⁶ SNPs of CADD score > 12.37 were considered potentially deleterious variants. Additionally, we performed a Bayesian colocalization analysis to identify loci that are common to both CVDs and OA. The posterior probability (PP) of five hypotheses was calculated, and genomic loci with PP.H4 value greater than 0.7 (indicating both traits are associated and share a single causal variant) were identified as colocalized loci with potential shared causal variants. To understand the shared mechanisms of the identified loci, we mapped nearby genes based on the lead SNPs within each locus. We also applied a generalized geneset analysis of GWAS data by using the MAGMA approach to determine the biological function of these pleiotropic loci, and a total of 10,678 gene sets, including curated gene sets (c2.all) and gene ontology (GO) terms (c5.bp, c5.cc, and c5.mf) from the Molecular Signatures Database (MSigDB), were tested. ²⁷ A Bonferroni correction was applied to all tested gene sets to minimize the risk of false positives, with a significance threshold of P < 0.05/10,678 = 4.68 × 10^–6. A genome-wide tissue-specific enrichment analysis was conducted using data from 54 GTEx tissues for the genome-wide pleiotropic results generated by PLACO. We also calculated the average expression (log2 transformed) of all identified pleiotropic genes across 54 GTEx tissues and assessed tissue specificity by examining differentially expressed genes in each tissue with up- and down-regulated DEGs precomputed based on the signs of the t-statistics. Metascape webtools (metascape.orgx) were used to perform a pathway enrichment analysis to ascertain the function of the mapped genes based on GO and DisGeNET database. Statistical significance of pathway enrichment was determined using P-values adjusted by the Bonferroni method to control for multiple comparisons (Bonferroni P<0.05 was considered significant).

2.10 Summary based Mendelian randomization (SMR)

SMR and HEIDI methods were used to test pleiotropic associations between gene expression levels and traits of interest based on summary-level data of GWAS and eQTL studies. ²⁸ For each probe, the top cis-QTL (within a 2 Mb window around the probe) was selected as the instrumental variable, requiring a QTL P value < 5.0×10^-8 and minor allele frequency (MAF) > 0.01. The SMR P values were adjusted for multiple testing using the false discovery rate (FDR). associations with SMR FDR < 0.05 and HEIDI>0.05 were considered significant.

2.11 Overview of drug targets and drugs under clinical trials

Therapeutic Target Database (TTD) and Open Targets database were used to search whether there were drugs aimed at certain pleiotropic genes that were in the clinical trial phase.^29,30 Druggable targets, drug names, indications, and max clinical trial phases were documented.

2.12 Multi-trait colocalization analysis

HyPrColoc method was used to perform multi-trait colocalization analysis among lipid/lipoprotein level, CVDs and OA. ³¹ Lipid/lipoprotein GWAS data contains triglyceride (TG) level (GCST90239664), total cholesterol (TC) level (GCST90239676), low density lipoprotein cholesterol (LDL-C) level (GCST90239658), high density lipoprotein cholesterol (HDL-C) levels (GCST90239652), and non-HDL cholesterol (non-HDL-C) levels (GCST90239670) from 1,320,016 European individuals, which could be downloaded from the GWAS Catalog database. We used the HyPrColoc R package with a regional window of 500 kb surrounding the pleiotropic loci. Posterior probability (PP) threshold of > 0.60 was applied to identify a high likelihood of a shared causal variant among the multiple traits.

2.13. Mendelian randomization analysis

We conducted a two-sample Mendelian randomization (MR) analysis to evaluate potential causal effects of CVDs on OA risk and OA on CVD risk reversely. ³² The “clumping” procedure of PLINK 1.9 software was applied to extract independent significant SNPs (P<5×10^-8), and r2 was set as 0.001 and window size was set as 10000 KB. Five MR approaches were used on each pair trait, including Inverse Variance Weighted, MR Egger Regression, Weighted Median, Simple Mode, and Weighted Mode. The instrumental SNPs were deposited into the IEUgwas and GWAS Catalog databases, and variables with potential confounding (smoking, blood lipid levels, inflammatory factor levels, metabolic diseases) were excluded.

Results

3.1 Characteristics of the study participants

In the final analytical samples, 2,103 individuals (34.3% of the 6,135 ELSA participants) and 2,871 individuals (19.3% of the 14,884 NHANES participants) were diagnosed with OA. Across both cohorts, OA patients shared a consistent clinical and demographic profile: they were significantly older and more likely to be female and white (or non-Hispanic white) compared to their non-OA counterparts. Regarding metabolic and lifestyle factors, OA participants in both studies presented with higher BMI, a greater prevalence of hypertension and diabetes, and distinct patterns in education, smoking status, and alcohol consumption. Furthermore, OA patients in both cohorts reported significantly higher frequencies of individual CVDs, including AP, MI, HF, and stroke, with CAD also being significantly more prevalent among OA patients in the NHANES study. Notably, OA participants in both cohorts were more likely to suffer from multiple concurrent cardiovascular conditions than those without OA (Tables 1 and 2).

Table 1.

Baseline characteristics of ELSA cohort grouped by OA condition.

​	Total (n = 6135)	Non-OA (n = 4032)	OA (n = 2103)	P value
Age(years)	68.777 ± 9.144	67.472 ± 9.156	71.278 ± 8.586	< 0.001
Sex (%)	​	​	​	< 0.001
Male	2704 (44.075)	1999 (49.578)	705 (33.524)	​
Female	3431 (55.925)	2033 (50.422)	1398 (66.476)	​
Level of education (%)	​	​	​	< 0.001
Basic	2277 (37.115)	1429 (35.441)	848 (40.323)	​
Intermediate	1786 (29.112)	1212 (30.06)	574 (27.294)	​
Higher	2072 (33.773)	1391 (34.499)	681 (32.382)	​
Race/ethnicity (%)	​	​	​	0.002
Non-white	166 (2.706)	128 (3.175)	38 (1.807)	​
White	5969 (97.294)	3904 (96.825)	2065 (98.193)	​
Body weight category (%)	​	​	​	< 0.001
Normal	2631 (42.885)	1874 (46.478)	757 (35.996)	​
Overweight	3274 (53.366)	2007 (49.777)	1267 (60.247)	​
Underweight	230 (3.749)	151 (3.745)	79 (3.757)	​
Smoking behavior (%)	​	​	​	< 0.001
Never	2356 (38.403)	1613 (40.005)	743 (35.33)	​
Ex_smoker	3232 (52.681)	2026 (50.248)	1206 (57.347)	​
Current	547 (8.916)	393 (9.747)	154 (7.323)	​
Alcohol consumption (%)	​	​	​	< 0.001
Never or seldom	1364 (22.233)	792 (19.643)	572 (27.199)	​
Occasional	2682 (43.716)	1785 (44.271)	897 (42.653)	​
Often	2089 (34.051)	1455 (36.086)	634 (30.147)	​
Hypertension (%)	​	​	​	< 0.001
No	3767 (61.402)	2633 (65.303)	1134 (53.923)	​
Yes	2368 (38.598)	1399 (34.697)	969 (46.077)	​
Diabetes (%)	​	​	​	0.008
No	5395 (87.938)	3578 (88.74)	1817 (86.4)	​
Yes	740 (12.062)	454 (11.26)	286 (13.6)	​
AP (%)	​	​	​	< 0.001
No	5942 (96.854)	3953 (98.041)	1989 (94.579)	​
Yes	193 (3.146)	79 (1.959)	114 (5.421)	​
MI (%)	​	​	​	0.011
No	5825 (94.947)	3849 (95.461)	1976 (93.961)	​
Yes	310 (5.053)	183 (4.539)	127 (6.039)	​
HF (%)	​	​	​	< 0.001
No	6081 (99.120)	4008 (99.405)	2073 (98.573)	​
Yes	54 (0.880)	24 (0.595)	30 (1.427)	​
Stroke (%)	​	​	​	< 0.001
No	5845 (95.273)	3890 (96.478)	1955 (92.962)	​
Yes	290 (4.727)	142 (3.522)	148 (7.038)	​
Concurrent CVD number (%)	​	​	​	< 0.001
0	5451 (88.851)	3677 (91.195)	1774 (84.356)	​
1	542 (8.835)	291 (7.217)	251 (11.935)	​
≥2	142 (2.315)	64 (1.587)	78 (3.709)	​

Table 2.

Baseline characteristics of NHANES cohort grouped by OA condition.

​	Total (n = 14884)	Non-OA(n = 12013)	OA (n = 2871)	P value
Age(years)	58.484 ± 12.192	56.914 ± 11.918	65.055 ± 11.084	< 0.001
Sex(%)	​	​	​	< 0.001
Male	7559 (50.786)	6511 (54.2)	1048 (36.503)	​
Female	7325 (49.214)	5502 (45.8)	1823 (63.497)	​
Race/ethnicity(%)	​	​	​	< 0.001
Non-Hispanic white	6898 (46.345)	5054 (42.071)	1844 (64.228)	​
Non-Hispanic black	3010 (20.223)	2556 (21.277)	454 (15.813)	​
Other race	4976 (33.432)	4403 (36.652)	573 (19.958)	​
Level of education(%)	​	​	​	< 0.001
Less than high school	1579 (10.609)	1381 (11.496)	198 (6.897)	​
High school	5284 (35.501)	4283 (35.653)	1001 (34.866)	​
More than high school	8021 (53.890)	6349 (52.851)	1672 (58.238)	​
PIR(%)	​	​	​	0.233
<1.3	3915 (26.303)	3190 (26.555)	725 (25.253)	​
1.3-3.5	5651 (37.967)	4526 (37.676)	1125 (39.185)	​
≥3.5	5318 (35.730)	4297 (35.77)	1021 (35.563)	​
BMI(kg/m2)	29.146 ± 6.351	28.732 ± 6.043	30.877 ± 7.252	< 0.001
Use of NSAIDs (%)	​	​	​	< 0.001
No	14051 (94.403)	11658 (97.045)	2393 (83.351)	​
Yes	833 (5.597)	355 (2.955)	478 (16.649)	​
Smoking behavior(%)	​	​	​	< 0.001
Never smoker	7846 (52.714)	6496 (54.075)	1350 (47.022)	​
Current smoker	2661 (17.878)	2232 (18.58)	429 (14.943)	​
Former smoker	4377 (29.407)	3285 (27.345)	1092 (38.036)	​
Alcohol consumption(%)	​	​	​	< 0.001
No	4895 (32.888)	3876 (32.265)	1019 (35.493)	​
Yes	9989 (67.112)	8137 (67.735)	1852 (64.507)	​
Hypertension(%)	​	​	​	< 0.001
No	9520 (63.961)	8214 (68.376)	1306 (45.489)	​
Yes	5364 (36.039)	3799 (31.624)	1565 (54.511)	​
Diabetes(%)	​	​	​	< 0.001
No	12551 (84.325)	10286 (85.624)	2265 (78.892)	​
Yes	2333 (15.675)	1727 (14.376)	606 (21.108)	​
AP(%)	​	​	​	< 0.001
No	14426 (96.923)	11740 (97.727)	2686 (93.556)	​
Yes	458 ( 3.077)	273 (2.273)	185 (6.444)	​
CAD(%)	​	​	​	< 0.001
No	14131 (94.941)	11520 (95.896)	2611 (90.944)	​
Yes	753 ( 5.059)	493 (4.104)	260 (9.056)	​
MI(%)	​	​	​	< 0.001
No	14173 (95.223)	11528 (95.963)	2645 (92.128)	​
Yes	711 ( 4.777)	485 (4.037)	226 (7.872)	​
HF(%)	​	​	​	< 0.001
No	14344 (96.372)	11663 (97.086)	2681 (93.382)	​
Yes	540 ( 3.628)	350 (2.914)	190 (6.618)	​
Stroke(%)	​	​	​	< 0.001
No	14249 (95.734)	11588 (96.462)	2661 (92.685)	​
Yes	635 ( 4.266)	425 (3.538)	210 (7.315)	​
Concurrent CVD number(%)	​	​	​	< 0.001
0	13014 (87.436)	10767 (89.628)	2247 (78.265)	​
1	1107 ( 7.438)	754 (6.277)	353 (12.295)	​
≥2	763 ( 5.126)	492 (4.096)	271 (9.439)	​

3.2 Association between OA and CVDs

In the ELSA cohort, logistic regression analysis demonstrated that OA was significantly associated with an increased risk of AP, HF, and stroke across all adjustment levels (Figure 3(a)). However, the positive association between OA and MI observed in the univariate model (Model 1) was attenuated and became non-significant after adjusting for covariates in Models 2 and 3. For the NHANES cohort, weighted logistic regression revealed that OA was positively associated with AP, CAD, MI, HF and stroke throughout all three models (Figure 3(a)). The risk of OA increased with the number of concurrent CVDs in both cohorts (Figure 3(b)). Notably, OA risk rose even without personal CVD history but with a family history of AP or MI in NHANES cohort, hinting at a genetic link between OA and certain CVDs (Figure 3(c)).

Figure 3.

Forest plots depicting OR (95% CI) in OA associated with different CVD conditions.

(A) Association of OA with 5 CVDs of the ELSA cohort (upper) and the NHANES cohort (lower). (B) Association of OA with number of concurrent CVDs of the ELSA cohort (upper) and the NHANES cohort (lower). (C) Association of OA with the presence of AP and/or MI in close relatives of participants without CVDs of the NHANES cohort. Note: For the ELSA cohort, Model 1 was with no adjustment; Model 2 was adjusted for age, sex, and race; Model 3 was adjusted for Model 2 plus education, body weight category, alcohol consumption, smoking behavior, hypertension, and diabetes. For the NHANES cohort, Model 1 was with no adjustment; Model 2 was adjusted for age, sex, and race; Model 3 was adjusted for Model 2 plus education, PIR, BMI, alcohol consumption, smoking behavior, hypertension, and diabetes. OA, osteoarthritis; AP, angina pectoris; CAD, cardiac artery disease; MI, myocardial infarction; HF, heart failure.

Subgroup analyses probed whether CVD-OA associations varied across population groups including age, sex, and body weight categories. In the ELSA cohort, significant interactions were found for gender in the AP-OA and the stroke-OA link (Supplementary Fig. S1-4). In the NHANES cohort, gender significantly interacted with AP, CAD, HF, and stroke associations with OA, while significant age interactions were observed for CAD, MI, and stroke (Supplementary Fig. S5-9). Across both cohorts, these significant CVD-OA associations were consistently stronger in females and those aged 40-60. While other subgroup interactions were not significant.

3.3 Genetic correlations between CVDs and OA

To explore whether the association between the prevalence of CVDs and OA was underpinned by shared genetic factors, we executed a comprehensive genome-wide pairwise trait pleiotropic analysis. Genome-wide correlations computed by LDSC suggested significant positive correlations between OA and 5 CVDs (Figure 4(a), Supplementary Table S2). The highest correlated traits are HF (rg=0.3094, p=5.06x 10^-14), followed by AP (rg=0.2229, p=6.11 x 10^-11), MI (rg=0.1685, p=1.65 x 10^-7), stroke (rg= 0.1605, p=1.28 x 10^-5), and CAD (rg=0.1530 p=1.22 x 10^-10). Moreover, results of high-definition likelihood (HDL) analysis were highly consistent with the LDSC outcome (Figure 4(b), Supplementary Table S2), indicating significant genetic correlations between CVDs and OA.

Figure 4.

Genetic correlation estimated by LDSC (A) and HDL (B) methods, respectively.

Note. Bonferroni-corrected significance threshold was set at P < 0.01 (0.05/5). LDSC, linkage disequilibrium score regression; HDL, high-definition likelihood; rg, genetic correlation; OA, osteoarthritis; AP, angina pectoris; CAD, cardiac artery disease; MI, myocardial infarction; HF, heart failure.

3.4 Shared loci between CVDs and OA

Using PLACO, we identified 3,628 SNPs (2,488 unique) associated with both CVDs and OA across five trait pairs, with no evidence of population stratification (Supplementary Fig. S10). FUMA analysis at P < 5 × 10^-8 detected 80 shared genomic risk loci (Figure 5(a), Supplementary Figs. S11-S12, Table S3). Of these, 46 SNPs (57.5%) showed consistent allelic effects, while 34 (42.5%) exhibited discordant associations (Supplementary Table S4). Colocalization analysis identified 20 loci with PP.H4 > 0.7 (e.g., 19q13.32, Supplementary Table S5), supported by regional plots (Supplementary Figs. S13-S17). Notably, loci like 2q33.2, 15q26.1, and 17p13.3 appeared in four trait pairs (Supplementary Table S6).

Figure 5.

Pleiotropic loci determination and gene-set and tissue-specific enrichment.

(A) Circular diagram displays pleiotropic loci determined by PLACO among five trait pairs. Bar plot of MAGMA gene-set (B) and tissue-specific (C) analysis for genome-wide pleiotropic results. Note: Shared loci identified by colocalization analysis are highlighted in green.

MAGMA functional annotation suggested these loci may influence cellular senescence, cholesterol metabolism, and bone remodeling (Figure 5(b), Supplementary Table S7), with shared pathways including response to endogenous stimulus and RUNX3 regulation across four trait pairs. Tissue enrichment analysis highlighted blood vessels (e.g., coronary artery, aorta) and reproductive organs (e.g., uterus, ovary) (Figure 5(c), Supplementary Table S8). ANNOVAR annotation revealed 39 lead SNPs (48.8%) were intronic, 20 (25.0%) intergenic, and 5 (6.3%) exonic.

3.5 Pleiotropic genes identified and enrichment analysis

Integration of SNP-level signals into gene-level associations identified 127 significant pleiotropic genes (84 unique) between CVDs and OA via MAGMA analysis (Figure 6(a), Supplementary Fig. S18). Among these, 28 genes (e.g., SMG6, UQCC1) showed pleiotropy across ≥3 trait pairs, with SMG6 and UQCC1 present in 4/5 pairs. eQTL analysis detected multiple hits for these genes in whole blood and arteries, with NBEAL1 and FES further validated by SMR. Figure 5(a) summarizes the pleiotropic gene landscape.

Figure 6.

Overview of pleiotropic genes (A) and drugs (B) targeting these genes for CVDs and OA.

Note. The signals represent hits of genes across different trait pairs. Drugs under clinical trials for endocrine disorders (ENDs, especially lipid metabolism related), CVDs or musculoskeletal disorders (MSD) treatment are displayed. eQTL expression quantitative trait loci, SMR summary-based Mendelian randomization, OA, osteoarthritis; AP, angina pectoris; CAD, cardiac artery disease; MI, myocardial infarction; HF, heart failure.

From the 84 pleiotropic genes, 16 druggable targets (e.g., FES, TGFB1) with clinical trial evidence were identified (Supplementary Table S9). Notably, 9 targets (e.g., FES, LPA) had drugs under clinical study for CVDs, musculoskeletal disorders (MSDs, including OA), and endocrine disorders (ENDs, especially lipid metabolism related) (Figure 6(b)). Among these, LPA was uniquely associated with CVD/END therapies, while others (e.g., FES, TGFB1) showed broader applicability.

Tissue enrichment revealed pleiotropic genes prominently expressed in tibial artery/aorta (NFU1, NBEAL1), adipose (MORF4L1, APOE), and reproductive tissues (FURIN, SMG6) (Supplementary Fig. S19). Significant SNP heritability enrichment was observed in tibial artery (all traits except HF) and heart, brain, pancreas, liver, and kidney cortex (Supplementary Figs. S20–S21, Table S10). GO analysis highlighted pathways like plasma lipoprotein remodeling and steroid hormone signaling (Figure 7(a)), while DisGeNET showed enrichment in CVDs, OA, and lipid metabolism (Figure 7(b)). PPI analysis identified two MCODE clusters: Cluster 1 (13 genes) linked to lipid dysregulation and OA, and Cluster 2 (9 genes) associated with CVDs (Figure 7(c)–(e)), implicating lipid dysregulation as a shared mechanism.​

Figure 7.

Pathway enrichment and PPI network of pleiotropic genes.

(A)Pathway enrichment for identified pleiotropic genes in GO database. (B) Pathway enrichment for identified pleiotropic genes in DisGeNET database. (C) PPI network of pleiotropic genes and 2 MCODE clusters. (D) DisGeNET enrichment of MCODE1 genes. (E) DisGeNET enrichment of MCODE2 genes. Note: lipid metabolism related processes and disorders are highlighted in brown.

3.6 Dysregulated lipid metabolism shared between CVDs and OA

Considering the findings from our functional enrichment and PPI analysis, which highlighted significant clustering in lipid-related pathways, we next aimed to clarify the specific lipid metabolic processes commonly affecting both CVDs and OA. Multi-trait colocalization analysis was performed among five lipid/lipoprotein traits (TC, TG, LDL-C, HDL-C, non-HDL-C), OA and five CVDs by using HyPrColoc (Table 3). Results highlight 13 pleiotropic loci, of which 7 were unique, and these loci support the important role of 4 unique lipid/lipoprotein traits in OA and CVDs. Notably, TC, LDL-C, and non-HDL-C were shared among 3 trait pairs (AP-OA, CAD-OA, and MI-OA), HDL-C was found in 2 trait pairs (CAD-OA and MI-OA), while TG was not significantly colocalized in any trait pair. These results indicated abnormal lipid (especially cholesterol) metabolism may potentially regulate disease progression in both CVDs and OA.

Table 3.

Multi-trait colocalization analysis highlighted key role of lipid metabolism.

Trait pairs	Lipid/lipoprotein type	Sample size	loci	candidate_snp	posterior_prob	regional_prob	posterior_explained_by_snp
OA-MI	Low density lipoprotein cholesterol levels	1320016 European	19:45411941:C:T	rs429358	0.8545	0.9352	1
OA-MI	Non-HDL cholesterol levels	1320016 European	19:45411941:C:T	rs429358	0.8545	0.9352	1
OA-MI	Total cholesterol levels	1320016 European	19:45411941:C:T	rs429358	0.8545	0.9352	1
OA-AP	Low density lipoprotein cholesterol levels	1320016 European	19:45412079:C:T	rs7412	0.8522	0.9352	1
OA-AP	Non-HDL cholesterol levels	1320016 European	19:45412079:C:T	rs7412	0.8522	0.9352	1
OA-AP	Total cholesterol levels	1320016 European	19:45412079:C:T	rs7412	0.8522	0.9352	1
OA-CAD	Low density lipoprotein cholesterol levels	1320016 European	19:45400747:A:T	rs61679753	0.8518	0.9352	1
OA-CAD	Non-HDL cholesterol levels	1320016 European	19:45400747:A:T	rs61679753	0.8518	0.9352	1
OA-CAD	Total cholesterol levels	1320016 European	19:45400747:A:T	rs61679753	0.8518	0.9352	1
OA-MI	High density lipoprotein cholesterol levels	1320016 European	3:136185680:A:G	rs6439640	0.7726	0.974	0.6505
OA-CAD	High density lipoprotein cholesterol levels	1320016 European	3:135925191:A:T	rs1154988	0.7491	0.9746	0.8079
OA-CAD	High density lipoprotein cholesterol levels	1320016 European	4:45179883:C:T	rs12641981	0.7341	0.7483	1
OA-MI	High density lipoprotein cholesterol levels	1320016 European	15:91437388:A:T	rs2521501	0.6475	0.8415	0.7715

Note. PP>0.6 were deemed significant. Abbreviations: PP, posterior probability. OA, osteoarthritis; AP, angina pectoris; CAD, coronary artery disease; MI, myocardial infarction; HF, heart failure.

3.7 The potential causal relationship between CVDs and OA estimated by MR

MR analysis was employed as a complementary approach to explore potential causal associations between OA and CVDs. Isolated estimates reached statistical significance in the Inverse Variance Weighted (IVW) analysis (negative associations of MI on OA risk, positive effect of OA on CAD, MI and stroke). However, these findings lacked cross-methodological consistency (Supplementary Fig. S22A-B). Besides these trait pairs, no other potential causal correlation was found.

Discussion

In the present research, we conducted an integrated analysis of the cross-sectional study based on nationally representative ELSA wave8 and NHANES 2005-2018 cohort with a genome-wide pleiotropic association study to explore the link between CVDs and OA. Our finding suggested compared with control group, CVD patients have a higher prevalence of OA. Meanwhile, we found significant genome-wide genetic correlations between them. Further analyses highlighted their pleiotropic genetic variants and loci, pleiotropic genes, and common biological pathways. All these findings supported strong epidemiologic and genetic relevance underlying these 2 types of diseases.

Emerging studies highlight a pronounced association between OA and several cardiovascular issues. In a large meta-analysis based on several epidemiological researches, a relative risk (RR) of 1.33 and 1.40 have been found for ischemic heart disease and HF in OA patients, respectively. ³³ Meanwhile, some follow-up surveys revealed OA was linked to an increased risk of CVD mortality.^33–35 Consistent with previous findings, our analysis based on the ELSA cohort and the NHANES cohort confirmed higher prevalence of common CVDs (AP, CAD, MI, HF, stroke) in OA patients. Regression analysis showed significant positive associations between most of these CVDs and OA after full adjustment, with risk increasing with CVD comorbidity count. Notably, this association persisted after age adjustment, suggesting age-independent mechanisms. Researchers are attempting to seek clues to their underlying connections besides age. For instance, chronic inflammation, a hallmark of OA, was also shown to promote the development of atherosclerosis and CVDs.^36,37 Some common lifestyle habits, such as physical inactivity and poor body weight management, can lead to increased mechanical stress on joints as well as contribute to vascular dysfunction.^38–40 Metabolic disorders, characterized by hypertension, dyslipidemia, and insulin resistance, simultaneously serve as catalysts exacerbating both OA and CVDs.^41–43 However, these factors may just represent shared phenotypes between the two diseases, rather than the initiating cause of their co-occurrence, and a deeper underlying mechanism needs to be clarified.

Considering cardiovascular system and articular structures (cartilage, bones, and synovial membranes) both originate from the mesoderm during embryonic development, it is plausible to deduce that certain shared genetic alterations might simultaneously lead to cardiovascular and articular disorders. ⁴⁴ Consistent with this hypothesis, we found an interesting phenomenon in our research: participants whose close relatives had AP or MI still showed higher risk of OA, even if they themselves had no cardiovascular issues, hinting at a potential genetic link between OA and specific CVDs. Multiple studies explored this link mainly through Mendelian randomization: Wang et al. found OA increased the incidence of ischemic stroke (OR=1.09) and CAD (OR=1.06). ¹⁶ MR study by Xu et al. found no evidence of genetically predicted association between OA and the risk of CAD or MI, however, they did report a suggestive protective association of MI (OR=0.95) and CAD (OR=0.90) with OA risk. ¹⁷ While IVW analysis of our study showed negative associations of MI on OA risk, positive effect of OA on CAD, MI, and stroke, they did not hold in pleiotropy-robust models (such as MR-Egger and Weighted Median). This methodological inconsistency points to potential horizontal pleiotropy or violations of other instrumental variable assumptions, indicating that a direct causal relationship between OA and CVDs cannot be simply inferred. To further clarify the genetic interplay underlying the co-occurrence of OA and CVDs, we adopted a different perspective to determine whether there is a shared genetic architecture between the two conditions. Through LDSC analysis, we provided compelling evidence for shared genetic connection between OA and CVDs, and the HDL analysis also indicates that individuals with OA should be aware of the potential for concurrent CVDs, which may partially elucidate why CVDs were considered as comorbidities of OA and vice versa.

We next identified a series of risk loci associated with both CVDs and OA, and some of them were observed in more than 1 phenotype pair (e.g., 7p21.1, 16q12.2). Among these risk loci, several loci are already proven to play an important role in the development of both CVDs and OA in previous literature. For example, loci on 16q12.2 (FTO) are positively associated with primary knee OA, which are also identified as susceptibility factors for CVDs such as myocardial fibrosis and heart failure.^45,46 HDAC9 (7p21.1), a key regulator promoting endothelial-mesenchymal transition and unfavorable atherosclerotic plaque phenotype, is regarded as a significant pathogenic gene in a wide range of CVDs (including CAD, MI, HF and stroke), with its SNPs being instrumental in the pathogenesis of these conditions, while recent large-scale GWAS data analysis also indicates the involvement of HDAC9 SNPs in OA progression, which is highly consistent with our results.^47,48 Besides these SNPs already reported in both types of diseases, our study further identified SNPs previously recognized only in the context of CVDs, such as FES (15q26.1), APOE (19q13.32), and ATP2B1 (12q21.33), had a significant correlation with OA as well. ⁴⁹ Similarly, loci like GDF5 (20q11.22), and TMEM129 (4p16.3) which were known to influence the risk of OA, have now been revealed to be pleiotropic loci with CVDs. ⁵⁰ Additionally, our research has pinpointed a multitude of previously unexplored pleiotropic variants in CVDs and OA, like NBEAL1 (2q33.2), GFPT1 (2p13.3), GNL3 (3p21.1), and SMG6 (17p13.3), which were further identified as significant eQTLs in whole blood and blood vessels by MAGMA and SMR analysis. Altogether, these novel loci, along with the previously established ones, may serve as potential diagnostic and druggable targets for CVDs and OA, emphasizing the necessity to fully understand their implications in disease management.

Shared genetic structures and pleiotropic gene annotation reveal common mechanisms between CVDs and OA.​​ Identified pleiotropic genes were enriched in CVD-related pathological changes (e.g., blood vessel remodeling) and OA phenotypes (e.g., chondrocyte differentiation, bone mineralization). They were also significantly annotated in shared pathogenic pathways, including VEGF receptor signaling, RUNX3-mediated CDKN1A transcription, and biosynthetic process regulation. Notably, lipid metabolism pathways (e.g., plasma lipoprotein remodeling, cholesterol measurement) were highly enriched, implicating these variants in disease onset/progression. Pleiotropy was significantly enriched in cardiovascular tissues (coronary artery, aorta, tibial artery) and reproductive organs (ovary, uterus, fallopian tube, cervix, testis), aligning with lesion sites and broader systemic involvement. A possible reason is that the reproductive system affects cardiovascular health and bone density/cartilage metabolism simultaneously through the secretion of sex hormones, such as estrogen and testosterone.^51,52 These findings collectively underscored a potential role played by lipid dysregulation, including lipid hormones (like sex hormones) dysregulation, in these two types of diseases. Supporting our findings, lipidomics revealed abnormal lipid metabolic patterns in both CVDs and OA, with certain lipids proven to regulate the progression of both conditions.^53,54 For example, cholesterol, notably LDL-C, contributes to atherosclerosis by plaque formation in arterial walls, potentially leading to heart attacks and strokes. ⁵⁵ Meanwhile disrupted cholesterol metabolism in chondrocytes results in cholesterol accumulation and extracellular matrix destruction. ⁵⁶ Our results demonstrated the critical influence of TC, LDL-C, and non-HDL-C in 3 trait pairs (AP-OA, CAD-OA, and MI-OA) and HDL-C in 2 trait pairs (CAD-OA and MI-OA), highlighting the importance of further investigation into lipid (especially cholesterol) metabolism as a potential therapeutic target for both CVDs and OA.

Limitations

However, the study has its limitations: First, we did not perform an a priori power analysis. While this poses a theoretical risk of underpowering in smaller studies, we mitigated this concern by maximizing our sample size of 21,019. Second, despite the substantial size of the OA GWAS summary data, the inclusion of 0.9% non-European individuals could potentially introduce a modest bias. Third, the tissue enrichment reference panels lacked joint-specific tissues and cells, resulting in the absence of OA related lesion sites in the enrichment outcomes, which could affect the comprehensiveness of our findings. Fourth, the identification of both OA and CVDs in the NHANES and ELSA cohorts relied entirely on self-reported physician diagnoses, which might introduce recall bias, and these surveys did not specify the anatomical site of OA (e.g., knee, hip), preventing us from performing subtype-specific analyses. Fifth, although we adjusted for the use of NSAIDs (a first-line OA treatment associated with cardiovascular events) in the NHANES cohort, the lack of medication records in the ELSA database precluded a similar adjustment. Finally, while our genetic analysis focused specifically on large, weight-bearing joints to reduce phenotypic heterogeneity, the GWAS summary data of pooling hip and knee as a single OA trait may still mask site-specific nuances in their association with CVDs.

Conclusions

In summary, we uncovered a significant positive correlation between the prevalence of OA and five CVDs (AP, CAD, MI, HF, and stroke). This may be attributed, at least in part, to the shared genetic predispositions between these conditions. We highlighted various pleiotropic loci and genes played a role in one or more trait pairs of CVDs and OA, like NBEAL1 (2q33.2), SMG6 (17p13.3), APOE (19q13.32), etc, and these variants may induce dysregulated lipid metabolism to promote their co-occurrence. Our research enhances the comprehension of the concurrent onset of OA and CVDs through the exploration of their shared genetic architecture. More precise diagnostic, therapeutic, and preventive measures for diseases may be developed based on these common functioning loci and genes.

ORCID iDs

Yuheng Lu https://orcid.org/0000-0002-2271-8748

Dexin Hu https://orcid.org/0009-0005-8979-623X

Hua Yuan https://orcid.org/0000-0001-7945-5136

Chenguang Zhao https://orcid.org/0000-0001-8986-5788

Ethical considerations

The Ethics Committee of Xijing Hospital granted approval for the study (KY20253289-1). This study involved secondary analyses of publicly available, de-identified datasets. Participation in NHANES and ELSA was voluntary, and written informed consent (including consent for future research studies) was obtained from all participants prior to the household interview.

Consent to participate

The GWAS summary statistics are derived from aggregated data and do not contain any individual-level information that could be used to identify participants. All data were accessed and used in accordance with the respective data use policies and end-user license agreements.

Data Availability Statement

All data generated or analyzed during this study are included in this published article and its supplementary information files.*